Files
dobetternorge-tools/includes
daveadmin f270a32056 fix(tools): tier-aware GPU cloud fallback for ask synthesis
When a persona-pinned GPU fine-tune is offline, degrade to the requested
quality tier's Bedrock model (Quick->Haiku, Pro->Sonnet) instead of a
hardcoded gpt-4o, so Pro genuinely differs from Quick while the pod is off.
Legacy/azure engines keep gpt-4o as the floor. Generalize the degraded
notice/trace wording (no longer asserts gpt-4o).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-06-15 18:50:39 +02:00
..
2026-05-23 10:17:34 +02:00
2026-05-23 10:17:34 +02:00